I call BS on this. For a LLM to recursively improve itself it would need to (small step) improve the training data and/or (big step) come up with fundamentally new architectures superior to transformers. The small step improvements might be doable. But nobody is making any claims about the big step improvements.

I have used custom code generators for years, generating 90+% of the code needed to write a typical biz application. Claude Code is useful and I use it every day. But it still hasn't beaten the productivity of my code generator.

Something I recently realized:

You can think of a view as a function that computes an output table from N input tables. Kinda like how Excel works but with tables instead of cells. That is extremely powerful.

Impressive work. It's nice to see LEAN being used for real-world algorithms.

Thanks! Yes, I hope AI and Lean will enable formally verified practical software.

> that has revolutionized several areas of computer science

That is quite a claim. I will argue that Categorical Databases most definitely haven't. Any areas where it is true?

Typescript is brilliant and should be carefully studied by anybody introducing a type system to a single typed ("untyped") language.

All good points. I have sent a link to this page to the author of the language (Panzerschrek).

I hope the author is ok with this kind of publicity. He might not have wanted attention if the language is not ready for public opinion.

IMHO science, technology, and economics has a much greater impact on history over the long term than any one individual.

An obvious example is the invention of the steam engine.

If I understand it correctly, more advanced logics (based around state transitions) doesn't have this problem (TLA+ etc.)

"The whipping will continue until morale improves"